Causality between credit depth and economic growth: evidence from 24 OECD countries

Earlier influential surveys include Levine (2005) and Ang (2008). Summarizing the literature up to the early 2000s, they identify the channels through which financial development facilitates growth. Levine (2005) notably focuses on specific functions such as producing ex ante information on investments and capital allocation, exerting corporate governance after providing finance, facilitating risk management, mobilizing savings, and easing the exchange of goods and services. When these functions are performed well, Levine argues, it increases the likelihood that financial development can be a robust predictor of future economic activity. Given that Granger causality hinges around the notion of predictability, it legitimizes the assessment of causal links between finance and growth.

Only papers published in refereed journals since 2008 are considered here. My search was conducted in the SCOPUS and ECONLIT databases. Though the choice of year is arbitrary, my hope was to select research with significant chronological overlap to the current study. Papers published before 2008 likely miss a significant part of the 1980–2013 observation period.

The sample includes Australia, Austria, Belgium, Chile, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Mexico, the Netherlands, Portugal, South Korea, Spain, Sweden, Switzerland, Turkey, the UK, and the USA. Canada, Luxemburg, New Zealand, and Norway are excluded due to significant data gaps in the series for domestic private credit to GDP.

As of early January 2016, there were 517 citations of the Kapetanios et al. (2003) paper in the SCOPUS database. To the best of my knowledge, their test has yet to be applied in a finance-growth context.

EViews 8 is used to perform the Zivot-Andrews and Perron tests; the CMR procedure is carried out in STATA 13. Technical details of the tests are omitted for brevity. Test output is available from the author upon request.

The tests of Levin et al. (2002) and Breitung (2002) assume a common unit root process. The test of Im et al. (2003), as well as ADF-Fisher and PP-Fisher tests, implies individual unit root processes. An overall judgment on the presence of group unit root is based on the summary of these test statistics.

CRED can equally be a dependent variable and GDP set as a predictor as the cointegrating relations are bidirectional.

In light of potential concerns with the over-parameterization of the VAR models, they have been re-estimated as Bayesian VARs (BVAR) with Litterman/Minnesota priors to check for the significance of the coefficients determining causal relationships in the VAR Granger causality tests. The results of the BVAR estimation validate the models in Table 1. Moreover, as the link between CRED and GDP for a given country may depend on the interactions with other countries’ variables, a global BVAR approach has also been adopted. It involves two steps. First, I assume that the relationship between CRED and GDP for a country in question depends on how SAV and TRADE variables for the selected country interact with CRED, GDP, SAV and TRADE for the rest 23 countries in the sample. Therefore, there are 94 variables that may affect the relationship between CRED and GDP for the selected country. To tackle the problem of dimensionality, standard factor analysis applies. It extracts 11 principal factors replacing the initial set of variables based on the Kaiser–Guttman procedure. This number of factors is the same across all countries in the sample. Second, I estimate 24 Bayesian VAR models with 13 variables (CRED and GDP for the country in question plus 11 factors). The results generally support the baseline conclusions: scarce evidence of causality between credit depth and growth and causality mainly running from CRED to GDP, if observed. Detailed output of these BVAR robustness checks is available from the author upon request.

Unlike the VAR Granger causality test, the SAV and TRADE variables cannot be endogenized for the Breitung–Candelon test. They are considered genuine conditioning variables, which explains the differences in the results of the two tests. For 16 of 24 countries, however, they coincide in the sense of causality presence and direction.

Ratios of stock market capitalization to GDP and stock market volume traded to GDP are available from 1988 for most sample countries. Reconstruction of annual data back to 1980 based on national historical financial statistics is not possible for some countries. For others, e.g., the UK and Sweden, the estimates are inconsistent with the post-1988 trajectories of these indicators.